News from the IT Home on October 3rd: Reuters published a lengthy article today (October 3rd) titled "Artificial Intelligence Competes to Change the World Before Running Out of Funds (AI's race to transform the world before the money runs out)." The article mentions that no new technology has ever attracted such a huge amount of funding as AI. Both the construction of railways and the rise of the internet sparked capital booms, but the scale of investment at that time has since been surpassed by AI.

PwC estimates that by 2050, the cumulative investment in data centers alone globally could exceed $30 trillion (Note from IT: The current exchange rate is approximately 201.43 trillion yuan), which is almost equivalent to the balance of U.S. national debt. Even after accounting for the impact of inflation, the scale of investment in AI infrastructure still far exceeds that during the railway construction boom and the internet bubble period.

At the same time, as one of the main participants in the AI competition, Anthropic plans to invest $518 billion (approximately 3.48 trillion yuan at current exchange rates) over the next few years. The prospectus shows that this amount exceeds 100 times Anthropic's revenue for the year 2025. Supporters believe that the changes brought about by AI could even surpass those brought by the invention of the steam engine and the industrialization it spurred.

Economists point out that the dazzling investment plans, massive expenditures, and high valuations of AI companies are all based on the expectation of a significant increase in productivity and substantial future profits. So far, there is almost no evidence or historical precedent to prove that these expectations will definitely come true.

JPMorgan Chase pointed out in August that in the United States, which is leading in the AI competition, a widespread increase in productivity “has yet to appear.” Therefore, whether the current valuations of AI companies can be maintained in the long term is also in question.

A study released by Bain last month suggests that relying solely on existing markets to improve productivity is not sufficient to support the current level of investment. New markets must emerge to fill the funding gap. These markets could range from robots controlled by AI to the development of new battery and semiconductor materials using AI.

Bain estimates that, including Google, Amazon, and Microsoft, the large-scale cloud service providers in the United States that are vigorously building infrastructure, as well as other companies in the AI sector, will need to generate over $4.2 trillion (approximately 28.2 trillion yuan at current exchange rates) in revenue over the next five years to afford this round of infrastructure expansion. The study points out that the question is whether applications that can recoup these investments will emerge in a timely manner.

Almost no one doubts that AI has the potential to transform various industries, from offices to scientific research laboratories. Past technological revolutions have shortened journeys that used to take days to just a few hours, and they have also enabled people to connect with the world with just a few keystrokes on a keyboard.

Investments ultimately require calculating returns, and loans also have repayment deadlines. Economists need to assess not only the ups and downs that investment booms will experience, but also what impact this round of AI investments will have on the global economy in the end.

JPMorgan Chase wrote: "Historical experience shows that prosperity driven by technology often ends when the infrastructure can no longer provide sufficient returns."

JPMorgan Chase used NVIDIA as an example for its calculations. NVIDIA’s chips constitute the core foundation of the AI industry. To support the company’s current valuation, the productivity in the United States needs to increase by 3% to 5% annually over the next 10 years. However, the Congressional Budget Office’s baseline forecast for productivity growth during the same period is only 1.75% per year.

According to some estimates, the United States accounts for about three-quarters of the global AI investment. Stein Van Nieuwbergen, an economist at Columbia Business School, estimates that just the AI investment in the United States alone between 2025 and 2032 could reach approximately 9 trillion US dollars (about 60.43 trillion yuan at current exchange rates), which is equivalent to 3.2% of the US GDP spent each year.

He estimates that if these investments are to achieve a 10% return rate, by 2032, the US AI industry will need to generate approximately $3.55 trillion (about 23.84 trillion yuan at current exchange rates) in revenue each year, yet the current revenue is only a fraction of this amount.

In a conference paper revised in October, Fanny Uveberg pointed out that the leverage ratio of debt financing for AI infrastructure is relatively high. Therefore, even a slight weakening in demand, project delays, or a decline in asset prices could result in much larger losses.

These dazzling numbers have not dampened the enthusiasm of American AI corporate leaders when discussing future changes.

Amodei once said that the future of AI could be "more beautiful than imagination." Oltman, on the other hand, believes that as models learn to improve themselves and make breakthroughs more quickly, the pace at which new miracles occur will be incredibly astonishing.

Google's Chief Strategy Officer, Jashmeet Sehgal, stated at a summit at the University of California, Berkeley in August that this "recursive self-improvement" ability that allows AI to continuously improve itself is an important part of the investment logic. Once achieved, productivity is expected to see an unprecedented increase.

Recursive self-improvement has the potential to exponentially enhance the capabilities of AI, but it also raises concerns about the risks to human survival: the speed of productivity improvement may not keep up with the financial time constraints of businesses.

Diane Coyle, an economist at the University of Cambridge in the UK, pointed out that when looking back at major technological revolutions in the past, the impact of new technologies on productivity usually takes about 10 to 50 years to fully manifest.

The economic team has estimated several possible economic growth scenarios that AI could bring about by 2030. Assuming there is no AI, the annual economic growth rate is 2%. If the impact of AI is minor, the growth rate is 2.4%; if the impact is significant, it rises to 5.4%; in extreme cases, it can reach 15.4%. The faster the economic growth, the more jobs will be lost, but it is not possible to determine the likelihood of each of these scenarios occurring.

Amodei predicted last year that AI could lead to the disappearance of half of junior white-collar jobs within 5 years. However, some researchers believe that so far, the most significant impact of AI on employment has been to make it more difficult for job seekers looking for office jobs to find positions.

Research in the United States and the United Kingdom shows that, against the backdrop of generally strong employment, recruitment by companies for new graduates in white-collar jobs, which AI are relatively adept at undertaking, has slowed down.

In August, researchers from Stanford University discovered that in professions significantly affected by AI, such as accounting and legal assistant roles, the employment rate for workers aged 22 to 25 was 19% lower than in less replaceable professions like cleaners and construction workers.

The report also points out that even if the changes brought about by AI do not occur as quickly as implied by corporate valuations and investment figures, the actual economic benefits that are generated will still remain. The financial panic of 1873 led to the bankruptcy of many railway tycoons, but railways did not disappear; similarly, after the bursting of the internet bubble in the 1990s, the internet continued to develop.